Re: [vox] [OT] unions

[Micah Cowan <MAPSmicah@cowanbox.com>]

On Fri, Aug 24, 2001 at 04:05:22PM -0700, Joel Baumert wrote:
> I think that unions had a purpose in the past, but government 
> regulation has made them less and less useful.  Normal everyday
> structures provide most members with enough portable benefits
> and security that I don't understand why this outdated 
> feature has not been banned! :-)
> 
> Joel

Are you kidding?  I sure hope so.

Unions provide the only completely portable and reliable method of
supporting polymorphic data structures.  Paraphrased real example:

struct common_event {
  int event_type;
  time_t event_time;
}

struct mouse_down_event {
  int event_type;
  time_t event_time;
  int x, y;
}

struct keyboard_event {
  int event_type;
  time_t event_time;
  int    char_code;
}

... etc...

union event {
  struct common_event     common;
  struct mouse_down_event m_event;
  struct keyboard_event   k_event;

   ....
}

If you have maybe 80 different event types, it'd be unreasonable to
store all the information in a struct.  The above "union event"
construct allows you to test the common portion of the structs via
.common.event_type, before deciding exactly what sort of event it is.
This sort of construct becomes absolutely crucial in object-oriented C
programming.

Note that you cannot portably (i.e., the C Standard does not define
the behavior if you do it) simply cast pointers to these structures
back and forth.  GTK+ does it this way, and AFAIK almost no compiler
actually refuses such a construct, but that doesn't mean future ones
won't.  As a principle, I always stray from the C standard as
absolutely seldom as possible (only in using system-specific
libraries, headers and features that are necessary).

Micah